🧑‍💼 Anthropic releases Cowork for Windows

Good Morning! Anthropic just brought Cowork (Claude Code for everyday work) to Windows, expanding Claude’s agent-style workflows beyond macOS. Plus, I’ll show you how to steal billion-dollar AI prompting techniques and use them yourself.

Plus, in today’s AI newsletter:

• Anthropic Releases Cowork on Windows

• LLaDA2.1 Breaks Autoregressive Norms

• Isomorphic Labs Breaks Past AlphaFold 3

• Steal Billion-Dollar AI Prompting Techniques

• 4 new AI tools worth trying

AI NEWS

🧑‍💼 Anthropic Releases Cowork on Windows

Cowork, Anthropic’s agentic workspace built on Claude Code, is now available on Windows with full feature parity. It’s designed to let Claude act more like a real coworker, planning, executing, and managing multi-step work across your files.

• Full Windows support with file access, plugins, MCP connectors, and multi-step tasks

• Global & folder instructions let you set tone, format, and work style once and reuse it

• Claude can read, edit, create, and organize files with higher autonomy than chat

• Available in research preview for Pro, Team, and Enterprise plans

Cowork signals Anthropic’s push beyond chat into true agentic workspaces. By giving Claude controlled access to files and long-running tasks, it moves AI closer to replacing fragmented tools with a single, persistent AI coworker, now accessible to Windows users too.

P.S. If you want to get in front of an audience of 20,000+ AI professionals and tech enthusiasts, get in touch with us here.

AI MODELS

🧩 LLaDA2.1 Breaks Autoregressive Norms

LLaDA2.1 is a 100B-parameter diffusion-based LLM that drafts fast, then self-corrects on the fly using Token-to-Token (T2T) editing, more like how humans write and revise.

• Hits 892 tokens/sec on HumanEval+ and 801 TPS on BigCodeBench

• Uses an Error-Correcting Editable (ECE) engine for real-time self-review

• Dual modes: Speedy Mode for rapid iteration, Quality Mode for precision

• Fully open-source with Day-0 SGLang support, production-ready today

Autoregressive models have dominated for years, but they can’t fix mistakes mid-generation. LLaDA2.1 shows that diffusion + self-editing can deliver both speed and accuracy at scale, putting “non-consensus” architectures back in the race.

AI NEWS

🧬 Isomorphic Labs Breaks Past AlphaFold 3

Isomorphic Labs introduced its Drug Design Engine (IsoDDE), a unified AI system that doesn’t just predict structures, but designs drugs end-to-end with unprecedented accuracy, speed, and cost efficiency.

• More than 2× accuracy gains over AlphaFold 3 on hard, novel protein, ligand tasks

• 2.3× better antibody predictions, including the notoriously hard CDR-H3 loop

• Binding-affinity predictions beat physics-based gold standards at a fraction of time and cost

• Discovers cryptic drug pockets from sequence alone, even in unseen biology

This moves drug discovery from years of trial-and-error toward faster, cheaper, computer-designed medicines, a real step toward fully AI-driven drug development.

HOW TO AI

🗂️ How to Steal Billion-Dollar AI Prompting Techniques and Use Them Yourself

In this tutorial, you’ll learn how to access leaked system prompts used by top-tier AI models like Claude, understand why they work so well, and reuse the same internal instructions to dramatically improve your own AI outputs. You don’t need to be a developer, and you don’t need paid tools, just copy, adapt, and apply.

🧰 Who is This For

• Creators who want AI responses that sound more human and structured

• Developers who want cleaner reasoning, better code, and fewer hallucinations

• Power users building custom GPTs, agents, or system prompts

• Anyone curious how billion-dollar AI models are actually instructed

STEP 1: Access the “Gold Mine” (The Prompt Repository)

In this tutorial, the first thing you’ll do is access the public GitHub repository that contains Claude’s internal system prompts. You don’t need coding skills for this.

Open the GitHub repository link in your browser. Once the page loads, you’ll see a list of files. Each file represents a different internal persona or task configuration used by Claude.

Look for files with names like claude_3_5_sonnet.md or claude_code.md. Click on any one of these files to open it. What you’re seeing now is the exact text Anthropic sends to Claude before you type your first message. This is essentially Claude’s “brain setup.”

STEP 2: “Secret Sauce” Inside the Prompt

The files may look long and intimidating, but you don’t need to understand every line. Focus on a few key sections that make the model feel intelligent and human-like.

Look for sections that describe how the model should think before answering. These are often labeled with things like reasoning blocks or internal thinking instructions. These sections guide the AI to break down problems step by step instead of rushing to an answer.

Next, notice the constraint instructions. These tell the AI what not to do, such as avoiding unnecessary apologies, not mentioning it is an AI, or staying concise. These constraints are a big reason Claude feels more natural and confident.

Finally, pay attention to tool usage instructions. These explain when and how the model should use tools like calculators, code editors, or structured outputs. This is what enables consistent, professional responses.

STEP 3: Build Your Own “Super Agent” System Prompt

Now comes the fun part. You don’t need to copy the entire file. Instead, copy the specific instruction patterns you like.

For example, if you like how Claude writes clean, readable code, copy the section that talks about clean code principles or structured problem solving. If you like its direct tone, copy the instruction that enforces clarity and brevity.

Paste these instructions at the very beginning of your own prompt in ChatGPT, Claude, or even a local model. Start with something like:
“Act as an expert assistant. Follow these internal instructions for your reasoning:” and then paste the selected instructions below.

You can tweak the language to match your use case, whether that’s coding, writing, research, or analysis.

STEP 4: Generate, Review, and Scale Up

To get the most value, turn these instructions into a permanent system prompt.

If you’re using tools like ChatGPT Custom GPTs or Claude Projects, open the Instructions or System Prompt section. Combine your favorite parts from multiple leaked prompts into one unified instruction set.

For example, you might use Claude’s directness rules to reduce fluff, its reasoning structure to improve clarity, and its tool-use logic to get better outputs for complex tasks.

Once saved, test it by asking a difficult question or giving a complex task. You’ll notice immediately that the AI now thinks, structures answers, and responds with the same professional discipline used by billion-dollar models.

At that point, you’re no longer just using AI, you’re controlling how it thinks.

xAI co-founder Jimmy Ba is leaving xAI, the sixth co-founder to depart; sources and social media posts: more than half a dozen researchers left in recent weeks.

OpenAI updates ChatGPT's deep research tool with GPT-5.2, a full-screen report view, and an option to focus research on specific websites.

GPT-5.3-Codex and Claude Opus 4.6 can handle the full app development lifecycle on their own, a sign of what's coming for most knowledge work within five years.

Global installed base of active smartphones grew 2% in 2025; eight OEMs had 200M+ active devices each, and nearly one in four active smartphones is an iPhone.

🧑‍💻 Claude Opus 4.6: Built for long, serious work in coding and research

⚙️ GPT-5.3 Codex: Faster coding model with strong reasoning

🤖 OpenAI Frontier: Enterprise platform for building and managing AI agents

🔎 Model Council: Perplexity tool to query multiple AI models at once

THAT’S IT FOR TODAY

Thanks for making it to the end! I put my heart into every email I send, I hope you are enjoying it. Let me know your thoughts so I can make the next one even better!

See you tomorrow :)

- Dr. Alvaro Cintas